Poetiq has officially established a new state-of-the-art on the ARC-AGI-2 Semi-Private Test Set, achieving a 54% success rate. This result was verified by the ARC Prize Team and represents the first time the system has broken through the 50% barrier for this benchmark.
- Poetiq's open-source ARC-AGI solver achieved 54% accuracy at a cost of $30.57 per problem.
- The previous best score was 45%, set by Gemini 3 Deep Think, which cost $77.16 per problem.
- The system utilizes Poetiq’s meta-system to optimize solutions using existing frontier models like Gemini 3 without requiring model fine-tuning.
- The meta-system learns from solved tasks to improve performance and can optimize AI components within larger systems.
This achievement demonstrates a new Pareto frontier for cost-effective reasoning, allowing the system to solve complex problems efficiently by leveraging existing model capabilities rather than building or training new ones.